Tesseract: PDF output without image

[olcc]

Hello,

I noticed the new "pdf" option in Tesseract, which creates a PDF file with the image and the background text. That's great !

But usually, the image given to Tesseract is not as nice as the starting image (because it is optimized for OCR, not for human visualization). Maybe it would be useful to provide the step before, i.e. the PDF of the generated text without the image, so that the user can paste it as a background text with pdftk for example.

[jbarlow83]

@olcc Tesseract is a raw OCR engine. Have a look at my project, OCRmyPDF, which provides a nice wrapper around Tesseract and takes care of many details to improve visualization.

[Wikinaut]

@olcc link is OCRmyPDF

[ws233]

@olcc, the way to produce PDF has significantly changed in Tesseract 8.04. So I have a plan to change this in future commits. I'll take your idea into consideration. But as I remeber the new implementation does not produce the text anymore. It outputs directly to the file. But even with such effort you are able read the file manually and modify as you wish.

[zdenop]

@olcc: tesseract puts to pdf image that you provided as input (e.g. file you see in pdf is not optimized for OCR as you claims). If you have another experience - please provide example. Otherwise close the "issue".

[olcc]

@jbarlow83: Thanks for pointing to the "OCRmyPDF" wrapper.
@ws233: Tesseract 8.04? I'm quite late, I only have 3.04! ;-) (from Debian)
@zdenop: Sorry, I didn't understand your message. Maybe my English is not good enough. My process is the following:
1) ORIGINAL.jpg -> OCR.tif (remove colors, apply threshold, etc.)
2) tesseract OCR.tif result -l eng pdf
If you say that showing OCR.tif in the PDF is the right thing to do, I disagree in general. I agree this is a very nice feature. However, most people want to have ORIGINAL.jpg with the ocr text.

[zdenop]

What I want to say is that if you run:
tesseract OCR.tif ORIGINAL pdf
than ORIGINAL.tif is included in ORIGINAL.pdf WITHOUT any modification. If you want to include ORIGINAL.jpg instead of OCR.tif than it is not tesseract issue ;-)

[Wikinaut]

@olcc we here fully rely on these "mixed-mode" PDFs as generated by

tesseract OCR.tif ORIGINAL pdf

which works with very high quality, depending on the quality what you input to tesseract. I hope, that the present "pdf" option ( -c tessedit_create_pdf=1 ) will really never be dropped from the code.

[amitdo]

@zdenop, is this functionality documented anywhere?

Could you point me to the exact place in the code where it's implemented?

[zdenop]

@amitdo: it is implemented in pdfrenderer

This is not real issue (no bug in tessseract), so I close this issue. Please use tesseract user forum for asking question/support.

[jbreiden]

ORIGINAL.tif is included in ORIGINAL.pdf WITHOUT any modification

Whenever possible. The design intent is to copy the image bytes without using a
decompress/compress whenever we can. Sometimes that is impossible (TIFF
is an enormously flexible graphics format) and sometimes we haven't quite
gotten there. For example, TIFF CCITT Group 4 still goes through a lossless
decompress/compress. Simply because we haven't done the work to optimize
this code path in Tesseract / Leptonica. All relevant Tesseract code is in
ai/pdfrenderer.cc but we try to push the image heavy lifting into Leptonica.

https://en.wikipedia.org/wiki/Tagged_Image_File_Format#TIFF_Compression_Tag

[sergiocallegari]

I'd like to support the original wish. Having something like

tesseract OCR.tif ORIGINAL pdf-overlay

to produce only the text overlay in a pdf file would provide a lot of flexibility. With this, you could write frontends to tesseract capable of overlaying the invisible text overlay on something different from OCR.tiff (e.g. a full color version of OCR.tif, etc.)